A Study On Electromagnetic Scattering Models For Random Rough Surfaces

Most of the simulation software for solving electromagnetic scattering problems can only be used for smooth surface,but the problem of electric field distribution in most complex scenes is based on random rough surface.And many scattering models for solving the complex field's distribution under high frequency have the limitations of low accuracy and poor material handling ability.In order to solve these problems,high frequency scattering models of rough surfaces are needed.Then,the high frequency scattering model of rough surface is used to realize the fast calculation of electromagnetic field distribution in typical scenes.The scattering model of this thesis has multiple inputs,which makes the model more complicated.In order to simplify the model,one-dimensional rough surface is chosen as the research basis.The correlation length,the root mean square height,the incident angle,the scattering angle and the polarization mode are used as the input parameters,and then the scattering values are obtained by the perturbation method and the method of moments.After the scattering coefficient is obtained,the scattering model can be established by a suitable algorithm.Then comparing the fitting value and the theoretical calculation value.There are two scattering models in this thesis:(1)The electromagnetic scattering model of one dimensional Gauss rough surface and exponential rough surface is established by least square method.The relationship function between scattering coefficient and incident angle,scattering angle,correlation length and root mean square height can be obtained.By comparing the fitting value with the calculated value,it is found that the electromagnetic scattering model obtained by the least square method has certain accuracy.(2)In order to get a more applicable and less error model,an electromagnetic scattering model is established by using BP neural network for one dimensional conductor/medium Gauss random rough surface.Then the relationship function of scattering coefficient with incident angle,scattering angle,correlation length,root mean square height and polarization mode is obtained.In the process of neural network modeling,two different output functions are selected,so two different models are obtained.By comparing the fitting values and calculated values of neural network,a more accurate neural network model is obtained.At the same time,the comparison results between the fitting values and the calculated values under different conditions are given to prove the reliability and accuracy of the fitting results of the BP neural network model.